EURO 2025 Leeds
Abstract Submission

2644. A Comprehensive Methodology for Predicting Volatility: Leveraging Intrinsic Entropy Estimates through the Integration of GARCH Models and LSTM Networks

Invited abstract in session TB-9: Complexity in finance, stream OR in Finance and Insurance .

Tuesday, 10:30-12:00
Room: Clarendon SR 2.01

Authors (first author is the speaker)

1. Claudiu Vinte
Economic Informatics and Cybernetics Department, Bucharest University of Economic Studies
2. Marcel Ausloos
School of Business
3. Bogdan Iftimie
Department of Applied Mathematics, Bucharest University of Economic Studies
4. Titus-Felix Furtună
Economic Informatics and Cybernetics Department, Bucharest University of Economic Studies

Abstract

To assess the overall volatility of the market, we utilize the Cross-Sectional Intrinsic Entropy (CSIE) model as the foundational methodology for monitoring the aggregate volatility of all symbols listed and traded on a specific stock market. In contrast, a market index intrinsic entropy (IE) volatility estimates are derived from n-day rolling windows. Consequently, CSIE market volatility estimates are incorporated into the research methodology as comparable n-day moving averages. We introduce a synergistic methodology that integrates GARCH models with Long Short-Term Memory (LSTM) networks, further augmented by intrinsic entropy estimates. This research employs over 5,500 historical end-of-day (EOD) data points, which encompass opening, high, low, and closing prices, along with traded volume (OHLCV) for all companies listed on the New York Stock Exchange (NYSE) and NASDAQ. Furthermore, it integrates the relevant market indices, specifically the S&P 500 and the NASDAQ Composite. To effectively model the relationship between consecutive values in a series of volatility estimates, we introduce specific step functions. The output produced by the proposed step functions is utilized by the LSTM networks. We conduct training for GARCH and LSTM models independently and subsequently integrate their predictions. By decoupling the models, this methodology facilitates the capture of distinct aspects of volatility and improves prediction accuracy for the step functions output.

Keywords

Status: accepted

Back to the list of papers